2 * Copyright (c) 2007-2011 Nicira Networks.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/system.h>
43 #include <asm/div64.h>
44 #include <linux/highmem.h>
45 #include <linux/netfilter_bridge.h>
46 #include <linux/netfilter_ipv4.h>
47 #include <linux/inetdevice.h>
48 #include <linux/list.h>
49 #include <linux/openvswitch.h>
50 #include <linux/rculist.h>
51 #include <linux/dmi.h>
52 #include <linux/workqueue.h>
53 #include <net/genetlink.h>
57 #include "vport-internal_dev.h"
62 * Writes to device state (add/remove datapath, port, set operations on vports,
63 * etc.) are protected by RTNL.
65 * Writes to other state (flow table modifications, set miscellaneous datapath
66 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
69 * Reads are protected by RCU.
71 * There are a few special cases (mostly stats) that have their own
72 * synchronization but they nest under all of above and don't interact with
76 /* Global list of datapaths to enable dumping them all out.
77 * Protected by genl_mutex.
79 static LIST_HEAD(dps
);
81 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
82 static void rehash_flow_table(struct work_struct
*work
);
83 static DECLARE_DELAYED_WORK(rehash_flow_wq
, rehash_flow_table
);
85 static struct vport
*new_vport(const struct vport_parms
*);
86 static int queue_gso_packets(int dp_ifindex
, struct sk_buff
*,
87 const struct dp_upcall_info
*);
88 static int queue_userspace_packet(int dp_ifindex
, struct sk_buff
*,
89 const struct dp_upcall_info
*);
91 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
92 static struct datapath
*get_dp(int dp_ifindex
)
94 struct datapath
*dp
= NULL
;
95 struct net_device
*dev
;
98 dev
= dev_get_by_index_rcu(&init_net
, dp_ifindex
);
100 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
109 /* Must be called with rcu_read_lock or RTNL lock. */
110 const char *ovs_dp_name(const struct datapath
*dp
)
112 struct vport
*vport
= rcu_dereference_rtnl(dp
->ports
[OVSP_LOCAL
]);
113 return vport
->ops
->get_name(vport
);
116 static int get_dpifindex(struct datapath
*dp
)
123 local
= rcu_dereference(dp
->ports
[OVSP_LOCAL
]);
125 ifindex
= local
->ops
->get_ifindex(local
);
134 static void destroy_dp_rcu(struct rcu_head
*rcu
)
136 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
138 ovs_flow_tbl_destroy((__force
struct flow_table
*)dp
->table
);
139 free_percpu(dp
->stats_percpu
);
143 /* Called with RTNL lock and genl_lock. */
144 static struct vport
*new_vport(const struct vport_parms
*parms
)
148 vport
= ovs_vport_add(parms
);
149 if (!IS_ERR(vport
)) {
150 struct datapath
*dp
= parms
->dp
;
152 rcu_assign_pointer(dp
->ports
[parms
->port_no
], vport
);
153 list_add(&vport
->node
, &dp
->port_list
);
159 /* Called with RTNL lock. */
160 void ovs_dp_detach_port(struct vport
*p
)
164 /* First drop references to device. */
166 rcu_assign_pointer(p
->dp
->ports
[p
->port_no
], NULL
);
168 /* Then destroy it. */
172 /* Must be called with rcu_read_lock. */
173 void ovs_dp_process_received_packet(struct vport
*p
, struct sk_buff
*skb
)
175 struct datapath
*dp
= p
->dp
;
176 struct sw_flow
*flow
;
177 struct dp_stats_percpu
*stats
;
178 struct sw_flow_key key
;
183 stats
= per_cpu_ptr(dp
->stats_percpu
, smp_processor_id());
185 /* Extract flow from 'skb' into 'key'. */
186 error
= ovs_flow_extract(skb
, p
->port_no
, &key
, &key_len
);
187 if (unlikely(error
)) {
193 flow
= ovs_flow_tbl_lookup(rcu_dereference(dp
->table
), &key
, key_len
);
194 if (unlikely(!flow
)) {
195 struct dp_upcall_info upcall
;
197 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
199 upcall
.userdata
= NULL
;
200 upcall
.pid
= p
->upcall_pid
;
201 ovs_dp_upcall(dp
, skb
, &upcall
);
203 stats_counter
= &stats
->n_missed
;
207 OVS_CB(skb
)->flow
= flow
;
209 stats_counter
= &stats
->n_hit
;
210 ovs_flow_used(OVS_CB(skb
)->flow
, skb
);
211 ovs_execute_actions(dp
, skb
);
214 /* Update datapath statistics. */
215 u64_stats_update_begin(&stats
->sync
);
217 u64_stats_update_end(&stats
->sync
);
220 static struct genl_family dp_packet_genl_family
= {
221 .id
= GENL_ID_GENERATE
,
222 .hdrsize
= sizeof(struct ovs_header
),
223 .name
= OVS_PACKET_FAMILY
,
224 .version
= OVS_PACKET_VERSION
,
225 .maxattr
= OVS_PACKET_ATTR_MAX
228 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
229 const struct dp_upcall_info
*upcall_info
)
231 struct dp_stats_percpu
*stats
;
235 if (upcall_info
->pid
== 0) {
240 dp_ifindex
= get_dpifindex(dp
);
246 if (!skb_is_gso(skb
))
247 err
= queue_userspace_packet(dp_ifindex
, skb
, upcall_info
);
249 err
= queue_gso_packets(dp_ifindex
, skb
, upcall_info
);
256 stats
= per_cpu_ptr(dp
->stats_percpu
, smp_processor_id());
258 u64_stats_update_begin(&stats
->sync
);
260 u64_stats_update_end(&stats
->sync
);
265 static int queue_gso_packets(int dp_ifindex
, struct sk_buff
*skb
,
266 const struct dp_upcall_info
*upcall_info
)
268 struct dp_upcall_info later_info
;
269 struct sw_flow_key later_key
;
270 struct sk_buff
*segs
, *nskb
;
273 segs
= skb_gso_segment(skb
, NETIF_F_SG
| NETIF_F_HW_CSUM
);
277 /* Queue all of the segments. */
280 err
= queue_userspace_packet(dp_ifindex
, skb
, upcall_info
);
284 if (skb
== segs
&& skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
) {
285 /* The initial flow key extracted by ovs_flow_extract()
286 * in this case is for a first fragment, so we need to
287 * properly mark later fragments.
289 later_key
= *upcall_info
->key
;
290 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
292 later_info
= *upcall_info
;
293 later_info
.key
= &later_key
;
294 upcall_info
= &later_info
;
296 } while ((skb
= skb
->next
));
298 /* Free all of the segments. */
306 } while ((skb
= nskb
));
310 static int queue_userspace_packet(int dp_ifindex
, struct sk_buff
*skb
,
311 const struct dp_upcall_info
*upcall_info
)
313 struct ovs_header
*upcall
;
314 struct sk_buff
*nskb
= NULL
;
315 struct sk_buff
*user_skb
; /* to be queued to userspace */
320 if (vlan_tx_tag_present(skb
)) {
321 nskb
= skb_clone(skb
, GFP_ATOMIC
);
325 nskb
= __vlan_put_tag(nskb
, vlan_tx_tag_get(nskb
));
333 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
338 len
= sizeof(struct ovs_header
);
339 len
+= nla_total_size(skb
->len
);
340 len
+= nla_total_size(FLOW_BUFSIZE
);
341 if (upcall_info
->cmd
== OVS_PACKET_CMD_ACTION
)
342 len
+= nla_total_size(8);
344 user_skb
= genlmsg_new(len
, GFP_ATOMIC
);
350 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
351 0, upcall_info
->cmd
);
352 upcall
->dp_ifindex
= dp_ifindex
;
354 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
355 ovs_flow_to_nlattrs(upcall_info
->key
, user_skb
);
356 nla_nest_end(user_skb
, nla
);
358 if (upcall_info
->userdata
)
359 nla_put_u64(user_skb
, OVS_PACKET_ATTR_USERDATA
,
360 nla_get_u64(upcall_info
->userdata
));
362 nla
= __nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, skb
->len
);
364 skb_copy_and_csum_dev(skb
, nla_data(nla
));
366 err
= genlmsg_unicast(&init_net
, user_skb
, upcall_info
->pid
);
373 /* Called with genl_mutex. */
374 static int flush_flows(int dp_ifindex
)
376 struct flow_table
*old_table
;
377 struct flow_table
*new_table
;
380 dp
= get_dp(dp_ifindex
);
384 old_table
= genl_dereference(dp
->table
);
385 new_table
= ovs_flow_tbl_alloc(TBL_MIN_BUCKETS
);
389 rcu_assign_pointer(dp
->table
, new_table
);
391 ovs_flow_tbl_deferred_destroy(old_table
);
395 static int validate_actions(const struct nlattr
*attr
,
396 const struct sw_flow_key
*key
, int depth
);
398 static int validate_sample(const struct nlattr
*attr
,
399 const struct sw_flow_key
*key
, int depth
)
401 const struct nlattr
*attrs
[OVS_SAMPLE_ATTR_MAX
+ 1];
402 const struct nlattr
*probability
, *actions
;
403 const struct nlattr
*a
;
406 memset(attrs
, 0, sizeof(attrs
));
407 nla_for_each_nested(a
, attr
, rem
) {
408 int type
= nla_type(a
);
409 if (!type
|| type
> OVS_SAMPLE_ATTR_MAX
|| attrs
[type
])
416 probability
= attrs
[OVS_SAMPLE_ATTR_PROBABILITY
];
417 if (!probability
|| nla_len(probability
) != sizeof(u32
))
420 actions
= attrs
[OVS_SAMPLE_ATTR_ACTIONS
];
421 if (!actions
|| (nla_len(actions
) && nla_len(actions
) < NLA_HDRLEN
))
423 return validate_actions(actions
, key
, depth
+ 1);
426 static int validate_set(const struct nlattr
*a
,
427 const struct sw_flow_key
*flow_key
)
429 const struct nlattr
*ovs_key
= nla_data(a
);
430 int key_type
= nla_type(ovs_key
);
432 /* There can be only one key in a action */
433 if (nla_total_size(nla_len(ovs_key
)) != nla_len(a
))
436 if (key_type
> OVS_KEY_ATTR_MAX
||
437 nla_len(ovs_key
) != ovs_key_lens
[key_type
])
441 const struct ovs_key_ipv4
*ipv4_key
;
443 case OVS_KEY_ATTR_PRIORITY
:
444 case OVS_KEY_ATTR_ETHERNET
:
447 case OVS_KEY_ATTR_IPV4
:
448 if (flow_key
->eth
.type
!= htons(ETH_P_IP
))
451 if (!flow_key
->ipv4
.addr
.src
|| !flow_key
->ipv4
.addr
.dst
)
454 ipv4_key
= nla_data(ovs_key
);
455 if (ipv4_key
->ipv4_proto
!= flow_key
->ip
.proto
)
458 if (ipv4_key
->ipv4_frag
!= flow_key
->ip
.frag
)
463 case OVS_KEY_ATTR_TCP
:
464 if (flow_key
->ip
.proto
!= IPPROTO_TCP
)
467 if (!flow_key
->ipv4
.tp
.src
|| !flow_key
->ipv4
.tp
.dst
)
472 case OVS_KEY_ATTR_UDP
:
473 if (flow_key
->ip
.proto
!= IPPROTO_UDP
)
476 if (!flow_key
->ipv4
.tp
.src
|| !flow_key
->ipv4
.tp
.dst
)
487 static int validate_userspace(const struct nlattr
*attr
)
489 static const struct nla_policy userspace_policy
[OVS_USERSPACE_ATTR_MAX
+ 1] = {
490 [OVS_USERSPACE_ATTR_PID
] = {.type
= NLA_U32
},
491 [OVS_USERSPACE_ATTR_USERDATA
] = {.type
= NLA_U64
},
493 struct nlattr
*a
[OVS_USERSPACE_ATTR_MAX
+ 1];
496 error
= nla_parse_nested(a
, OVS_USERSPACE_ATTR_MAX
,
497 attr
, userspace_policy
);
501 if (!a
[OVS_USERSPACE_ATTR_PID
] ||
502 !nla_get_u32(a
[OVS_USERSPACE_ATTR_PID
]))
508 static int validate_actions(const struct nlattr
*attr
,
509 const struct sw_flow_key
*key
, int depth
)
511 const struct nlattr
*a
;
514 if (depth
>= SAMPLE_ACTION_DEPTH
)
517 nla_for_each_nested(a
, attr
, rem
) {
518 /* Expected argument lengths, (u32)-1 for variable length. */
519 static const u32 action_lens
[OVS_ACTION_ATTR_MAX
+ 1] = {
520 [OVS_ACTION_ATTR_OUTPUT
] = sizeof(u32
),
521 [OVS_ACTION_ATTR_USERSPACE
] = (u32
)-1,
522 [OVS_ACTION_ATTR_PUSH_VLAN
] = sizeof(struct ovs_action_push_vlan
),
523 [OVS_ACTION_ATTR_POP_VLAN
] = 0,
524 [OVS_ACTION_ATTR_SET
] = (u32
)-1,
525 [OVS_ACTION_ATTR_SAMPLE
] = (u32
)-1
527 const struct ovs_action_push_vlan
*vlan
;
528 int type
= nla_type(a
);
530 if (type
> OVS_ACTION_ATTR_MAX
||
531 (action_lens
[type
] != nla_len(a
) &&
532 action_lens
[type
] != (u32
)-1))
536 case OVS_ACTION_ATTR_UNSPEC
:
539 case OVS_ACTION_ATTR_USERSPACE
:
540 err
= validate_userspace(a
);
545 case OVS_ACTION_ATTR_OUTPUT
:
546 if (nla_get_u32(a
) >= DP_MAX_PORTS
)
551 case OVS_ACTION_ATTR_POP_VLAN
:
554 case OVS_ACTION_ATTR_PUSH_VLAN
:
556 if (vlan
->vlan_tpid
!= htons(ETH_P_8021Q
))
558 if (!(vlan
->vlan_tci
& htons(VLAN_TAG_PRESENT
)))
562 case OVS_ACTION_ATTR_SET
:
563 err
= validate_set(a
, key
);
568 case OVS_ACTION_ATTR_SAMPLE
:
569 err
= validate_sample(a
, key
, depth
);
585 static void clear_stats(struct sw_flow
*flow
)
589 flow
->packet_count
= 0;
590 flow
->byte_count
= 0;
593 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
595 struct ovs_header
*ovs_header
= info
->userhdr
;
596 struct nlattr
**a
= info
->attrs
;
597 struct sw_flow_actions
*acts
;
598 struct sk_buff
*packet
;
599 struct sw_flow
*flow
;
607 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
608 !a
[OVS_PACKET_ATTR_ACTIONS
] ||
609 nla_len(a
[OVS_PACKET_ATTR_PACKET
]) < ETH_HLEN
)
612 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
613 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
617 skb_reserve(packet
, NET_IP_ALIGN
);
619 memcpy(__skb_put(packet
, len
), nla_data(a
[OVS_PACKET_ATTR_PACKET
]), len
);
621 skb_reset_mac_header(packet
);
622 eth
= eth_hdr(packet
);
624 /* Normally, setting the skb 'protocol' field would be handled by a
625 * call to eth_type_trans(), but it assumes there's a sending
626 * device, which we may not have. */
627 if (ntohs(eth
->h_proto
) >= 1536)
628 packet
->protocol
= eth
->h_proto
;
630 packet
->protocol
= htons(ETH_P_802_2
);
632 /* Build an sw_flow for sending this packet. */
633 flow
= ovs_flow_alloc();
638 err
= ovs_flow_extract(packet
, -1, &flow
->key
, &key_len
);
642 err
= ovs_flow_metadata_from_nlattrs(&flow
->key
.phy
.priority
,
643 &flow
->key
.phy
.in_port
,
644 a
[OVS_PACKET_ATTR_KEY
]);
648 err
= validate_actions(a
[OVS_PACKET_ATTR_ACTIONS
], &flow
->key
, 0);
652 flow
->hash
= ovs_flow_hash(&flow
->key
, key_len
);
654 acts
= ovs_flow_actions_alloc(a
[OVS_PACKET_ATTR_ACTIONS
]);
658 rcu_assign_pointer(flow
->sf_acts
, acts
);
660 OVS_CB(packet
)->flow
= flow
;
661 packet
->priority
= flow
->key
.phy
.priority
;
664 dp
= get_dp(ovs_header
->dp_ifindex
);
670 err
= ovs_execute_actions(dp
, packet
);
687 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
688 [OVS_PACKET_ATTR_PACKET
] = { .type
= NLA_UNSPEC
},
689 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
690 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
693 static struct genl_ops dp_packet_genl_ops
[] = {
694 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
695 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
696 .policy
= packet_policy
,
697 .doit
= ovs_packet_cmd_execute
701 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
)
704 struct flow_table
*table
= genl_dereference(dp
->table
);
706 stats
->n_flows
= ovs_flow_tbl_count(table
);
708 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
709 for_each_possible_cpu(i
) {
710 const struct dp_stats_percpu
*percpu_stats
;
711 struct dp_stats_percpu local_stats
;
714 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
717 start
= u64_stats_fetch_begin_bh(&percpu_stats
->sync
);
718 local_stats
= *percpu_stats
;
719 } while (u64_stats_fetch_retry_bh(&percpu_stats
->sync
, start
));
721 stats
->n_hit
+= local_stats
.n_hit
;
722 stats
->n_missed
+= local_stats
.n_missed
;
723 stats
->n_lost
+= local_stats
.n_lost
;
727 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
728 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
729 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
730 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
733 static struct genl_family dp_flow_genl_family
= {
734 .id
= GENL_ID_GENERATE
,
735 .hdrsize
= sizeof(struct ovs_header
),
736 .name
= OVS_FLOW_FAMILY
,
737 .version
= OVS_FLOW_VERSION
,
738 .maxattr
= OVS_FLOW_ATTR_MAX
741 static struct genl_multicast_group ovs_dp_flow_multicast_group
= {
742 .name
= OVS_FLOW_MCGROUP
745 /* Called with genl_lock. */
746 static int ovs_flow_cmd_fill_info(struct sw_flow
*flow
, struct datapath
*dp
,
747 struct sk_buff
*skb
, u32 pid
,
748 u32 seq
, u32 flags
, u8 cmd
)
750 const int skb_orig_len
= skb
->len
;
751 const struct sw_flow_actions
*sf_acts
;
752 struct ovs_flow_stats stats
;
753 struct ovs_header
*ovs_header
;
759 sf_acts
= rcu_dereference_protected(flow
->sf_acts
,
760 lockdep_genl_is_held());
762 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
766 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
768 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
770 goto nla_put_failure
;
771 err
= ovs_flow_to_nlattrs(&flow
->key
, skb
);
774 nla_nest_end(skb
, nla
);
776 spin_lock_bh(&flow
->lock
);
778 stats
.n_packets
= flow
->packet_count
;
779 stats
.n_bytes
= flow
->byte_count
;
780 tcp_flags
= flow
->tcp_flags
;
781 spin_unlock_bh(&flow
->lock
);
784 NLA_PUT_U64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
));
787 NLA_PUT(skb
, OVS_FLOW_ATTR_STATS
,
788 sizeof(struct ovs_flow_stats
), &stats
);
791 NLA_PUT_U8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, tcp_flags
);
793 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
794 * this is the first flow to be dumped into 'skb'. This is unusual for
795 * Netlink but individual action lists can be longer than
796 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
797 * The userspace caller can always fetch the actions separately if it
798 * really wants them. (Most userspace callers in fact don't care.)
800 * This can only fail for dump operations because the skb is always
801 * properly sized for single flows.
803 err
= nla_put(skb
, OVS_FLOW_ATTR_ACTIONS
, sf_acts
->actions_len
,
805 if (err
< 0 && skb_orig_len
)
808 return genlmsg_end(skb
, ovs_header
);
813 genlmsg_cancel(skb
, ovs_header
);
817 static struct sk_buff
*ovs_flow_cmd_alloc_info(struct sw_flow
*flow
)
819 const struct sw_flow_actions
*sf_acts
;
822 sf_acts
= rcu_dereference_protected(flow
->sf_acts
,
823 lockdep_genl_is_held());
825 /* OVS_FLOW_ATTR_KEY */
826 len
= nla_total_size(FLOW_BUFSIZE
);
827 /* OVS_FLOW_ATTR_ACTIONS */
828 len
+= nla_total_size(sf_acts
->actions_len
);
829 /* OVS_FLOW_ATTR_STATS */
830 len
+= nla_total_size(sizeof(struct ovs_flow_stats
));
831 /* OVS_FLOW_ATTR_TCP_FLAGS */
832 len
+= nla_total_size(1);
833 /* OVS_FLOW_ATTR_USED */
834 len
+= nla_total_size(8);
836 len
+= NLMSG_ALIGN(sizeof(struct ovs_header
));
838 return genlmsg_new(len
, GFP_KERNEL
);
841 static struct sk_buff
*ovs_flow_cmd_build_info(struct sw_flow
*flow
,
843 u32 pid
, u32 seq
, u8 cmd
)
848 skb
= ovs_flow_cmd_alloc_info(flow
);
850 return ERR_PTR(-ENOMEM
);
852 retval
= ovs_flow_cmd_fill_info(flow
, dp
, skb
, pid
, seq
, 0, cmd
);
857 static int ovs_flow_cmd_new_or_set(struct sk_buff
*skb
, struct genl_info
*info
)
859 struct nlattr
**a
= info
->attrs
;
860 struct ovs_header
*ovs_header
= info
->userhdr
;
861 struct sw_flow_key key
;
862 struct sw_flow
*flow
;
863 struct sk_buff
*reply
;
865 struct flow_table
*table
;
871 if (!a
[OVS_FLOW_ATTR_KEY
])
873 error
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
877 /* Validate actions. */
878 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
879 error
= validate_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &key
, 0);
882 } else if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
) {
887 dp
= get_dp(ovs_header
->dp_ifindex
);
892 table
= genl_dereference(dp
->table
);
893 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
895 struct sw_flow_actions
*acts
;
897 /* Bail out if we're not allowed to create a new flow. */
899 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_SET
)
902 /* Expand table, if necessary, to make room. */
903 if (ovs_flow_tbl_need_to_expand(table
)) {
904 struct flow_table
*new_table
;
906 new_table
= ovs_flow_tbl_expand(table
);
907 if (!IS_ERR(new_table
)) {
908 rcu_assign_pointer(dp
->table
, new_table
);
909 ovs_flow_tbl_deferred_destroy(table
);
910 table
= genl_dereference(dp
->table
);
915 flow
= ovs_flow_alloc();
917 error
= PTR_ERR(flow
);
923 /* Obtain actions. */
924 acts
= ovs_flow_actions_alloc(a
[OVS_FLOW_ATTR_ACTIONS
]);
925 error
= PTR_ERR(acts
);
927 goto error_free_flow
;
928 rcu_assign_pointer(flow
->sf_acts
, acts
);
930 /* Put flow in bucket. */
931 flow
->hash
= ovs_flow_hash(&key
, key_len
);
932 ovs_flow_tbl_insert(table
, flow
);
934 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
938 /* We found a matching flow. */
939 struct sw_flow_actions
*old_acts
;
940 struct nlattr
*acts_attrs
;
942 /* Bail out if we're not allowed to modify an existing flow.
943 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
944 * because Generic Netlink treats the latter as a dump
945 * request. We also accept NLM_F_EXCL in case that bug ever
949 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
&&
950 info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
| NLM_F_EXCL
))
953 /* Update actions. */
954 old_acts
= rcu_dereference_protected(flow
->sf_acts
,
955 lockdep_genl_is_held());
956 acts_attrs
= a
[OVS_FLOW_ATTR_ACTIONS
];
958 (old_acts
->actions_len
!= nla_len(acts_attrs
) ||
959 memcmp(old_acts
->actions
, nla_data(acts_attrs
),
960 old_acts
->actions_len
))) {
961 struct sw_flow_actions
*new_acts
;
963 new_acts
= ovs_flow_actions_alloc(acts_attrs
);
964 error
= PTR_ERR(new_acts
);
965 if (IS_ERR(new_acts
))
968 rcu_assign_pointer(flow
->sf_acts
, new_acts
);
969 ovs_flow_deferred_free_acts(old_acts
);
972 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
973 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
976 if (a
[OVS_FLOW_ATTR_CLEAR
]) {
977 spin_lock_bh(&flow
->lock
);
979 spin_unlock_bh(&flow
->lock
);
984 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
985 ovs_dp_flow_multicast_group
.id
, info
->nlhdr
,
988 netlink_set_err(init_net
.genl_sock
, 0,
989 ovs_dp_flow_multicast_group
.id
, PTR_ERR(reply
));
998 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1000 struct nlattr
**a
= info
->attrs
;
1001 struct ovs_header
*ovs_header
= info
->userhdr
;
1002 struct sw_flow_key key
;
1003 struct sk_buff
*reply
;
1004 struct sw_flow
*flow
;
1005 struct datapath
*dp
;
1006 struct flow_table
*table
;
1010 if (!a
[OVS_FLOW_ATTR_KEY
])
1012 err
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
1016 dp
= get_dp(ovs_header
->dp_ifindex
);
1020 table
= genl_dereference(dp
->table
);
1021 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
1025 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
1026 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
1028 return PTR_ERR(reply
);
1030 return genlmsg_reply(reply
, info
);
1033 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1035 struct nlattr
**a
= info
->attrs
;
1036 struct ovs_header
*ovs_header
= info
->userhdr
;
1037 struct sw_flow_key key
;
1038 struct sk_buff
*reply
;
1039 struct sw_flow
*flow
;
1040 struct datapath
*dp
;
1041 struct flow_table
*table
;
1045 if (!a
[OVS_FLOW_ATTR_KEY
])
1046 return flush_flows(ovs_header
->dp_ifindex
);
1047 err
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
1051 dp
= get_dp(ovs_header
->dp_ifindex
);
1055 table
= genl_dereference(dp
->table
);
1056 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
1060 reply
= ovs_flow_cmd_alloc_info(flow
);
1064 ovs_flow_tbl_remove(table
, flow
);
1066 err
= ovs_flow_cmd_fill_info(flow
, dp
, reply
, info
->snd_pid
,
1067 info
->snd_seq
, 0, OVS_FLOW_CMD_DEL
);
1070 ovs_flow_deferred_free(flow
);
1072 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1073 ovs_dp_flow_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1077 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1079 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1080 struct datapath
*dp
;
1081 struct flow_table
*table
;
1083 dp
= get_dp(ovs_header
->dp_ifindex
);
1087 table
= genl_dereference(dp
->table
);
1090 struct sw_flow
*flow
;
1093 bucket
= cb
->args
[0];
1095 flow
= ovs_flow_tbl_next(table
, &bucket
, &obj
);
1099 if (ovs_flow_cmd_fill_info(flow
, dp
, skb
,
1100 NETLINK_CB(cb
->skb
).pid
,
1101 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1102 OVS_FLOW_CMD_NEW
) < 0)
1105 cb
->args
[0] = bucket
;
1111 static struct genl_ops dp_flow_genl_ops
[] = {
1112 { .cmd
= OVS_FLOW_CMD_NEW
,
1113 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1114 .policy
= flow_policy
,
1115 .doit
= ovs_flow_cmd_new_or_set
1117 { .cmd
= OVS_FLOW_CMD_DEL
,
1118 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1119 .policy
= flow_policy
,
1120 .doit
= ovs_flow_cmd_del
1122 { .cmd
= OVS_FLOW_CMD_GET
,
1123 .flags
= 0, /* OK for unprivileged users. */
1124 .policy
= flow_policy
,
1125 .doit
= ovs_flow_cmd_get
,
1126 .dumpit
= ovs_flow_cmd_dump
1128 { .cmd
= OVS_FLOW_CMD_SET
,
1129 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1130 .policy
= flow_policy
,
1131 .doit
= ovs_flow_cmd_new_or_set
,
1135 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1136 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1137 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1140 static struct genl_family dp_datapath_genl_family
= {
1141 .id
= GENL_ID_GENERATE
,
1142 .hdrsize
= sizeof(struct ovs_header
),
1143 .name
= OVS_DATAPATH_FAMILY
,
1144 .version
= OVS_DATAPATH_VERSION
,
1145 .maxattr
= OVS_DP_ATTR_MAX
1148 static struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
1149 .name
= OVS_DATAPATH_MCGROUP
1152 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1153 u32 pid
, u32 seq
, u32 flags
, u8 cmd
)
1155 struct ovs_header
*ovs_header
;
1156 struct ovs_dp_stats dp_stats
;
1159 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_datapath_genl_family
,
1164 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1167 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1170 goto nla_put_failure
;
1172 get_dp_stats(dp
, &dp_stats
);
1173 NLA_PUT(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
), &dp_stats
);
1175 return genlmsg_end(skb
, ovs_header
);
1178 genlmsg_cancel(skb
, ovs_header
);
1183 static struct sk_buff
*ovs_dp_cmd_build_info(struct datapath
*dp
, u32 pid
,
1186 struct sk_buff
*skb
;
1189 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1191 return ERR_PTR(-ENOMEM
);
1193 retval
= ovs_dp_cmd_fill_info(dp
, skb
, pid
, seq
, 0, cmd
);
1196 return ERR_PTR(retval
);
1201 /* Called with genl_mutex and optionally with RTNL lock also. */
1202 static struct datapath
*lookup_datapath(struct ovs_header
*ovs_header
,
1203 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1205 struct datapath
*dp
;
1207 if (!a
[OVS_DP_ATTR_NAME
])
1208 dp
= get_dp(ovs_header
->dp_ifindex
);
1210 struct vport
*vport
;
1213 vport
= ovs_vport_locate(nla_data(a
[OVS_DP_ATTR_NAME
]));
1214 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1217 return dp
? dp
: ERR_PTR(-ENODEV
);
1220 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1222 struct nlattr
**a
= info
->attrs
;
1223 struct vport_parms parms
;
1224 struct sk_buff
*reply
;
1225 struct datapath
*dp
;
1226 struct vport
*vport
;
1230 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1235 if (!try_module_get(THIS_MODULE
))
1236 goto err_unlock_rtnl
;
1239 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1241 goto err_put_module
;
1242 INIT_LIST_HEAD(&dp
->port_list
);
1244 /* Allocate table. */
1246 rcu_assign_pointer(dp
->table
, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS
));
1250 dp
->stats_percpu
= alloc_percpu(struct dp_stats_percpu
);
1251 if (!dp
->stats_percpu
) {
1253 goto err_destroy_table
;
1256 /* Set up our datapath device. */
1257 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1258 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1259 parms
.options
= NULL
;
1261 parms
.port_no
= OVSP_LOCAL
;
1262 parms
.upcall_pid
= nla_get_u32(a
[OVS_DP_ATTR_UPCALL_PID
]);
1264 vport
= new_vport(&parms
);
1265 if (IS_ERR(vport
)) {
1266 err
= PTR_ERR(vport
);
1270 goto err_destroy_percpu
;
1273 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1274 info
->snd_seq
, OVS_DP_CMD_NEW
);
1275 err
= PTR_ERR(reply
);
1277 goto err_destroy_local_port
;
1279 list_add_tail(&dp
->list_node
, &dps
);
1282 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1283 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1287 err_destroy_local_port
:
1288 ovs_dp_detach_port(rtnl_dereference(dp
->ports
[OVSP_LOCAL
]));
1290 free_percpu(dp
->stats_percpu
);
1292 ovs_flow_tbl_destroy(genl_dereference(dp
->table
));
1296 module_put(THIS_MODULE
);
1303 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1305 struct vport
*vport
, *next_vport
;
1306 struct sk_buff
*reply
;
1307 struct datapath
*dp
;
1311 dp
= lookup_datapath(info
->userhdr
, info
->attrs
);
1316 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1317 info
->snd_seq
, OVS_DP_CMD_DEL
);
1318 err
= PTR_ERR(reply
);
1322 list_for_each_entry_safe(vport
, next_vport
, &dp
->port_list
, node
)
1323 if (vport
->port_no
!= OVSP_LOCAL
)
1324 ovs_dp_detach_port(vport
);
1326 list_del(&dp
->list_node
);
1327 ovs_dp_detach_port(rtnl_dereference(dp
->ports
[OVSP_LOCAL
]));
1329 /* rtnl_unlock() will wait until all the references to devices that
1330 * are pending unregistration have been dropped. We do it here to
1331 * ensure that any internal devices (which contain DP pointers) are
1332 * fully destroyed before freeing the datapath.
1336 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1337 module_put(THIS_MODULE
);
1339 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1340 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1350 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1352 struct sk_buff
*reply
;
1353 struct datapath
*dp
;
1356 dp
= lookup_datapath(info
->userhdr
, info
->attrs
);
1360 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1361 info
->snd_seq
, OVS_DP_CMD_NEW
);
1362 if (IS_ERR(reply
)) {
1363 err
= PTR_ERR(reply
);
1364 netlink_set_err(init_net
.genl_sock
, 0,
1365 ovs_dp_datapath_multicast_group
.id
, err
);
1369 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1370 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1376 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1378 struct sk_buff
*reply
;
1379 struct datapath
*dp
;
1381 dp
= lookup_datapath(info
->userhdr
, info
->attrs
);
1385 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1386 info
->snd_seq
, OVS_DP_CMD_NEW
);
1388 return PTR_ERR(reply
);
1390 return genlmsg_reply(reply
, info
);
1393 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1395 struct datapath
*dp
;
1396 int skip
= cb
->args
[0];
1399 list_for_each_entry(dp
, &dps
, list_node
) {
1402 if (ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).pid
,
1403 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1404 OVS_DP_CMD_NEW
) < 0)
1414 static struct genl_ops dp_datapath_genl_ops
[] = {
1415 { .cmd
= OVS_DP_CMD_NEW
,
1416 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1417 .policy
= datapath_policy
,
1418 .doit
= ovs_dp_cmd_new
1420 { .cmd
= OVS_DP_CMD_DEL
,
1421 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1422 .policy
= datapath_policy
,
1423 .doit
= ovs_dp_cmd_del
1425 { .cmd
= OVS_DP_CMD_GET
,
1426 .flags
= 0, /* OK for unprivileged users. */
1427 .policy
= datapath_policy
,
1428 .doit
= ovs_dp_cmd_get
,
1429 .dumpit
= ovs_dp_cmd_dump
1431 { .cmd
= OVS_DP_CMD_SET
,
1432 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1433 .policy
= datapath_policy
,
1434 .doit
= ovs_dp_cmd_set
,
1438 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1439 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1440 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1441 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1442 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1443 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1444 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1447 static struct genl_family dp_vport_genl_family
= {
1448 .id
= GENL_ID_GENERATE
,
1449 .hdrsize
= sizeof(struct ovs_header
),
1450 .name
= OVS_VPORT_FAMILY
,
1451 .version
= OVS_VPORT_VERSION
,
1452 .maxattr
= OVS_VPORT_ATTR_MAX
1455 struct genl_multicast_group ovs_dp_vport_multicast_group
= {
1456 .name
= OVS_VPORT_MCGROUP
1459 /* Called with RTNL lock or RCU read lock. */
1460 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1461 u32 pid
, u32 seq
, u32 flags
, u8 cmd
)
1463 struct ovs_header
*ovs_header
;
1464 struct ovs_vport_stats vport_stats
;
1467 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_vport_genl_family
,
1472 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1474 NLA_PUT_U32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
);
1475 NLA_PUT_U32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
);
1476 NLA_PUT_STRING(skb
, OVS_VPORT_ATTR_NAME
, vport
->ops
->get_name(vport
));
1477 NLA_PUT_U32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, vport
->upcall_pid
);
1479 ovs_vport_get_stats(vport
, &vport_stats
);
1480 NLA_PUT(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1483 err
= ovs_vport_get_options(vport
, skb
);
1484 if (err
== -EMSGSIZE
)
1487 return genlmsg_end(skb
, ovs_header
);
1492 genlmsg_cancel(skb
, ovs_header
);
1496 /* Called with RTNL lock or RCU read lock. */
1497 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 pid
,
1500 struct sk_buff
*skb
;
1503 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1505 return ERR_PTR(-ENOMEM
);
1507 retval
= ovs_vport_cmd_fill_info(vport
, skb
, pid
, seq
, 0, cmd
);
1510 return ERR_PTR(retval
);
1515 /* Called with RTNL lock or RCU read lock. */
1516 static struct vport
*lookup_vport(struct ovs_header
*ovs_header
,
1517 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1519 struct datapath
*dp
;
1520 struct vport
*vport
;
1522 if (a
[OVS_VPORT_ATTR_NAME
]) {
1523 vport
= ovs_vport_locate(nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1525 return ERR_PTR(-ENODEV
);
1527 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1528 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1530 if (port_no
>= DP_MAX_PORTS
)
1531 return ERR_PTR(-EFBIG
);
1533 dp
= get_dp(ovs_header
->dp_ifindex
);
1535 return ERR_PTR(-ENODEV
);
1537 vport
= rcu_dereference_rtnl(dp
->ports
[port_no
]);
1539 return ERR_PTR(-ENOENT
);
1542 return ERR_PTR(-EINVAL
);
1545 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1547 struct nlattr
**a
= info
->attrs
;
1548 struct ovs_header
*ovs_header
= info
->userhdr
;
1549 struct vport_parms parms
;
1550 struct sk_buff
*reply
;
1551 struct vport
*vport
;
1552 struct datapath
*dp
;
1557 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1558 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1562 dp
= get_dp(ovs_header
->dp_ifindex
);
1567 if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1568 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1571 if (port_no
>= DP_MAX_PORTS
)
1574 vport
= rtnl_dereference(dp
->ports
[port_no
]);
1579 for (port_no
= 1; ; port_no
++) {
1580 if (port_no
>= DP_MAX_PORTS
) {
1584 vport
= rtnl_dereference(dp
->ports
[port_no
]);
1590 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1591 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1592 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1594 parms
.port_no
= port_no
;
1595 parms
.upcall_pid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1597 vport
= new_vport(&parms
);
1598 err
= PTR_ERR(vport
);
1602 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1604 if (IS_ERR(reply
)) {
1605 err
= PTR_ERR(reply
);
1606 ovs_dp_detach_port(vport
);
1609 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1610 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1618 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1620 struct nlattr
**a
= info
->attrs
;
1621 struct sk_buff
*reply
;
1622 struct vport
*vport
;
1626 vport
= lookup_vport(info
->userhdr
, a
);
1627 err
= PTR_ERR(vport
);
1632 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1633 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
)
1636 if (!err
&& a
[OVS_VPORT_ATTR_OPTIONS
])
1637 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1638 if (!err
&& a
[OVS_VPORT_ATTR_UPCALL_PID
])
1639 vport
->upcall_pid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1641 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1643 if (IS_ERR(reply
)) {
1644 err
= PTR_ERR(reply
);
1645 netlink_set_err(init_net
.genl_sock
, 0,
1646 ovs_dp_vport_multicast_group
.id
, err
);
1650 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1651 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1658 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1660 struct nlattr
**a
= info
->attrs
;
1661 struct sk_buff
*reply
;
1662 struct vport
*vport
;
1666 vport
= lookup_vport(info
->userhdr
, a
);
1667 err
= PTR_ERR(vport
);
1671 if (vport
->port_no
== OVSP_LOCAL
) {
1676 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1678 err
= PTR_ERR(reply
);
1682 ovs_dp_detach_port(vport
);
1684 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1685 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1692 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1694 struct nlattr
**a
= info
->attrs
;
1695 struct ovs_header
*ovs_header
= info
->userhdr
;
1696 struct sk_buff
*reply
;
1697 struct vport
*vport
;
1701 vport
= lookup_vport(ovs_header
, a
);
1702 err
= PTR_ERR(vport
);
1706 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1708 err
= PTR_ERR(reply
);
1714 return genlmsg_reply(reply
, info
);
1721 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1723 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1724 struct datapath
*dp
;
1728 dp
= get_dp(ovs_header
->dp_ifindex
);
1733 for (port_no
= cb
->args
[0]; port_no
< DP_MAX_PORTS
; port_no
++) {
1734 struct vport
*vport
;
1736 vport
= rcu_dereference(dp
->ports
[port_no
]);
1740 if (ovs_vport_cmd_fill_info(vport
, skb
, NETLINK_CB(cb
->skb
).pid
,
1741 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1742 OVS_VPORT_CMD_NEW
) < 0)
1747 cb
->args
[0] = port_no
;
1753 static void rehash_flow_table(struct work_struct
*work
)
1755 struct datapath
*dp
;
1759 list_for_each_entry(dp
, &dps
, list_node
) {
1760 struct flow_table
*old_table
= genl_dereference(dp
->table
);
1761 struct flow_table
*new_table
;
1763 new_table
= ovs_flow_tbl_rehash(old_table
);
1764 if (!IS_ERR(new_table
)) {
1765 rcu_assign_pointer(dp
->table
, new_table
);
1766 ovs_flow_tbl_deferred_destroy(old_table
);
1772 schedule_delayed_work(&rehash_flow_wq
, REHASH_FLOW_INTERVAL
);
1775 static struct genl_ops dp_vport_genl_ops
[] = {
1776 { .cmd
= OVS_VPORT_CMD_NEW
,
1777 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1778 .policy
= vport_policy
,
1779 .doit
= ovs_vport_cmd_new
1781 { .cmd
= OVS_VPORT_CMD_DEL
,
1782 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1783 .policy
= vport_policy
,
1784 .doit
= ovs_vport_cmd_del
1786 { .cmd
= OVS_VPORT_CMD_GET
,
1787 .flags
= 0, /* OK for unprivileged users. */
1788 .policy
= vport_policy
,
1789 .doit
= ovs_vport_cmd_get
,
1790 .dumpit
= ovs_vport_cmd_dump
1792 { .cmd
= OVS_VPORT_CMD_SET
,
1793 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1794 .policy
= vport_policy
,
1795 .doit
= ovs_vport_cmd_set
,
1799 struct genl_family_and_ops
{
1800 struct genl_family
*family
;
1801 struct genl_ops
*ops
;
1803 struct genl_multicast_group
*group
;
1806 static const struct genl_family_and_ops dp_genl_families
[] = {
1807 { &dp_datapath_genl_family
,
1808 dp_datapath_genl_ops
, ARRAY_SIZE(dp_datapath_genl_ops
),
1809 &ovs_dp_datapath_multicast_group
},
1810 { &dp_vport_genl_family
,
1811 dp_vport_genl_ops
, ARRAY_SIZE(dp_vport_genl_ops
),
1812 &ovs_dp_vport_multicast_group
},
1813 { &dp_flow_genl_family
,
1814 dp_flow_genl_ops
, ARRAY_SIZE(dp_flow_genl_ops
),
1815 &ovs_dp_flow_multicast_group
},
1816 { &dp_packet_genl_family
,
1817 dp_packet_genl_ops
, ARRAY_SIZE(dp_packet_genl_ops
),
1821 static void dp_unregister_genl(int n_families
)
1825 for (i
= 0; i
< n_families
; i
++)
1826 genl_unregister_family(dp_genl_families
[i
].family
);
1829 static int dp_register_genl(void)
1836 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
1837 const struct genl_family_and_ops
*f
= &dp_genl_families
[i
];
1839 err
= genl_register_family_with_ops(f
->family
, f
->ops
,
1846 err
= genl_register_mc_group(f
->family
, f
->group
);
1855 dp_unregister_genl(n_registered
);
1859 static int __init
dp_init(void)
1861 struct sk_buff
*dummy_skb
;
1864 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > sizeof(dummy_skb
->cb
));
1866 pr_info("Open vSwitch switching datapath\n");
1868 err
= ovs_flow_init();
1872 err
= ovs_vport_init();
1874 goto error_flow_exit
;
1876 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
1878 goto error_vport_exit
;
1880 err
= dp_register_genl();
1882 goto error_unreg_notifier
;
1884 schedule_delayed_work(&rehash_flow_wq
, REHASH_FLOW_INTERVAL
);
1888 error_unreg_notifier
:
1889 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1898 static void dp_cleanup(void)
1900 cancel_delayed_work_sync(&rehash_flow_wq
);
1902 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
1903 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1908 module_init(dp_init
);
1909 module_exit(dp_cleanup
);
1911 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1912 MODULE_LICENSE("GPL");